Anda di halaman 1dari 9

Ind. Eng. Chem. Res.

1997, 36, 4299-4307 4299

A Continuous Model for C7+ Fraction Characterization of Petroleum

Mohammad R. Riazi†
Chemical Engineering Department, Kuwait University, P.O. Box 5969, Safat 13060, Kuwait

In this paper, based on a two-parameter distribution model, a method is presented to predict

complete property distributions for molecular weight, boiling point, specific gravity, and refractive
index of a C7+ fraction. Only three mixture bulk properties such as molecular weight, specific
gravity, and refractive index are needed for this technique. If a distillation (TBP) analysis for
the fraction is available, then only two bulk properties such as molecular weight and specific
gravity (or refractive index) are sufficient. Predicted distributions for various properties are
compared with experimental data of some 48 crude samples. The method is also applied for
flash vaporization of a Russian crude oil, and predicted distributions for feed, vapor, and liquid
streams are compared with actual data. Splitting and lumping schemes of petroleum fractions
using the proposed distribution model are also presented by two different aproaches with detailed
computational procedures.

Introduction Unfortunately, complete experimental data on the

distribution of molecular weight, TBP, and density for
Characterization of crude oil is an important step in a hydrocarbon-plus fraction are seldom available. Usu-
the application of equations of state for pressure- ally bulk properties such as molecular weight, specific
volume-temperature (PVT) predictions and phase be- gravity, density, or refractive index are available. In
havior calculations. Such calculations are needed in some cases only complete TBP analysis may be available
reservoir simulation or in the design and operation of for a hydrocarbon-plus fraction. The main purpose of
refinery distillation columns. An insufficient description this paper was to present a new method that can be
of heavier hydrocarbons (e.g., heptanes and heavier; used to predict complete ditributions for various proper-
C7+) reduces the accuracy of PVT predictions as shown ties of a C7+ fraction based on the knowledge of a
by Whitson (1983). Volumetric and phase behaviors of mixture’s bulk properties which are easily measurable.
even volatile oil and gas-condensate samples are quite Another objective of this work was to develop splitting
sensitive to properties of the heavier components. In and lumping schemes of petroleum fractions based on
the determination of properties of a C7+ fraction, usually the proposed distribution model.
distributions of some basic properties such as molecular
weight, true boiling point (TBP), and specific gravity (or Two-Parameter Distribution Model
density) are needed. Application of continuous distribu-
tion models in the prediction of the phase behavior of Riazi (1989) developed a simple and versatile distri-
heavy reservoir fluids is also demonstrated by Kawan- bution model for various properties of a hydrocarbon-
ada et al. (1991). Whitson (1983, 1984) used a gamma plus fraction in the following form:
distribution model to describe molecular weight/mole
[BA ln(x*1 )]
fraction relations for C7+ fractions. Another widely used P* ) (1)
distribution model is the exponential function as dis-
cussed by Ahmed (1989). Gamma distribution model
is a three-parameter function while the exponential where P* ) (P - P0)/P0 and x* ) 1 - x in which x is the
function is a two-parameter model. In addition, the cumulative weight, mole, or volume fraction and P is a
gamma and exponential models are mainly used for the property such as absolute boiling point (Tb), molecular
molecular weight distribution of plus fractions. A weight (M), specific gravity (S), density (d), or refractive
method based on a constant Watson K factor is outlined index parameter (I). Parameter I is related to the
by Soreide (1989) and Whitson (1984) to generate the refractive index (n) at 20 °C by the equation:
distribution of specific gravity. When the distribution
model is known, then a lumped fraction can be spilt into n2 - 1
several pseudocompounds or single carbon number I) (2)
n2 + 2
(SCN) groups with known mole fractions by using
mathematical techniques such as the Gaussian quadra-
P0 is the parameter specific for each property (T0, M0,
ture method (Stroud and Secrest, 1966). Soreide (1989)
S0, I0) and each sample. A is also a parameter specific
has shown the application of the Gaussian quadrature
for each property (AT, AM, AS, AI) and each sample. B is
method in obtaining appropriate pseudocomponents for
a parameter specific for each property (BT, BM, BS, BI)
a C7+ fraction using the gamma distribution model.
but is the same for all samples. It was determined that
Once pseudocomponents are determined, then correla-
for most samples examined BS ) BI ) 3, BM ) 1, and
tions such as Riazi and Daubert (1980) or Riazi and Al-
BT ) 1.5. In eq 1, for specific gravity (S), density (d),
Sahhaf (1996) can be used to estimate various properties and refractive index parameter (I), x is the cumulative
of a C7+ fraction (Ahmed, 1989). volume or weight fraction, while for molecular weight
(M), x is the cumulative mole fraction.

Fax: (+965) 4839498. Telephone: (+965) 4817662. E- With values of B known for different properties, eq 1
mail: becomes a two-parameter distribution model where P0
S0888-5885(97)00260-1 CCC: $14.00 © 1997 American Chemical Society
4300 Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997

and A should be determined. As discussed by Riazi Table 1. Bulk Properties for C7+ Samples Used in This
(1989), because eq 1 well represents fractions containing Study
heavy components and it is applicable to all various data fluid residue
parameters (M, Tb, S, I, d) with only two parameters to sample sourcea typeb M7+ S7+ n7+
be determined, it is an appropriate model to be used in 1 a BO 142.8 0.7717 1.4274
this study. Equation 1 can be converted into a linear 2 a BO 193.3 0.7995 1.4345
form: 3 b BO 289.4 0.9041 1.5194
4 b BO 206.2 0.8505 1.4512
Y ) C1 + C2X (3) 5 b BO 348.2 0.9386 1.5057
6 c BO 177.5 0.8067 1.4472
7 d BO 227.5 0.8518 1.4752
where: Y ) ln P*, X ) ln ln(1/x*), B ) 1/C2, and A ) 8 e BO 257.2 0.8732 1.4584
B exp(C1B). P0 and A can be determined by linear 9 e BO 205.7 0.8412 1.4457
regression of data through eq 3. The probability density 10 f BO 256.1 0.8759 1.4962
function F(P*) deduced from eq 1 is in the following 11 g BO 210.9 0.8451 1.4708
form: 12 g BO 217.1 0.8449 1.4723
13 g BO 236.2 0.8525 1.4814
14 g BO 232.0 0.8658 1.4839
B2 B-1 B
F(P*) )
P* exp - P*B
A ( ) (4) 15
17 h GC 149.4 0.8195 1.4594
This equation can also be written in terms of property, 18 h GC 120.2 0.7955 1.4440
P, if P* is replaced by its definition. Average mixture 19 h GC 118.9 0.7597 1.4234
20 i GC 149.9 0.8165 1.4595
properties calculated from eq 1 (or eq 4) in terms of P*
21 i GC 150.5 0.8162 1.4813
are given by 22 i GC 155.4 0.8086 1.4497
23 j GC 129.1 0.7865 1.4387
∫0∞P*F(P*) dP* ) (BA) 1
P*av ) (
Γ1+ )
(5) 24
where Γ is the gamma function. Equation 5 can be 27 j GC 153.7 0.7871 1.4281
applied to M, Tb, S, and I. For S and I, distributions 28 j GC 176.5 0.8043 1.4334
29 j GC 175.2 0.8033 1.4470
must be in cumulative volume or weight fractions. 30 j BO 216.7 0.8517 1.4526
However, as is discussed by Riazi (1989), if S is given 31 j BO 232.9 0.8534 1.4769
in terms of cumulative weight fraction, then S*av is 32 j BO 242.9 0.8833 1.4980
calculated by the relation 33 j BO 211.4 0.8458 1.4694
34 j BO 211.2 0.8459 1.4714

∫0∞F(S*) S*dS*
35 j BO 247.9 0.8839 1.4946
av ) (6) 36 j BO 221.5 0.8338 1.4625
+1 37 j GC 125.3 0.7737 1.4302
38 j GC 132.1 0.8039 1.4469
If fixed values for parameter B are used for different 39 j GC 128.4 0.7909 1.4401
properties, then eq 5 gives the following relations for 40 j BO 267.8 0.9033 1.4735
the average properties: 41 j GC 172.0 0.8127 1.4544
42 j GC 151.6 0.7917 1.4400
M*av ) AM (7) 43 j GC 122.5 0.7633 1.4250
44 j GC 140.1 0.8068 1.4358
45 j GC 170.5 0.8113 1.4537
T*av ) 0.689AT2/3 (8) 46 j GC 175.2 0.8055 1.4366
47 j GC 119.3 0.7838 1.4379
av ) 0.619AI
I* (9) 48 j GC 163.8 0.8137 1.4573
a References are as follows: (a) Jacoby and Berry (1959); (b)
Once P*av has been determined, the average property Lee et al. (1979); (c) Austad et al. (1983); (d) Hariu and Sage (1969);
Pav for the mixture can be obtained from the relation (e) Haaland (1981); (f) Berge (1981); (g) Pedersen et al. (1984a);
(h) Pedersen et al. (1984b); (i) Pedersen et al. (1985); (j) Whitson
Pav ) P0(1 + P* et al. (1988). b BO: black oil. GC: gas condensate.
av) (10)

For the specific gravity, S, when it is represented in bulk properties for M, S, and n for each sample of C7+
terms of cumulative weight fraction, then the average are given in Table 1. When eq 1 is used to predict
value is calculated from the following relations: distributions of various properties for samples listed in
Table 1, average absolute deviations (AAD) of 3 for M,
if AS > 0.05 3 K for Tb, and 0.6% for both S and n were obtained for
a total of 650 subfractions. The gamma distribution
Sav ) S0(1.3818 + 0.3503AS - 0.1932AS2 + model predicts molecular weight and boiling point
distributions with accuracy nearly the same as eq 1
0.059 ln AS) (11) while the constant Watson K method predicts specific
gravity distributions with AAD of 1.2%. The exponen-
if AS e 0.05 tial model predicts molecular weight with AAD of 4. The
exponential distribution model is not suitable to repre-
Sav ) S0(1.25355 + 1.44886AS - 5.9777AS2 + sent Tb, S, and n distributions.
0.02951 ln AS) (12) A comparison between various distribution models
when applied to a heavy fraction (Rodgers et al., 1987)
For 48 samples of oils and condensates listed in Table is shown in Figure 1. The mixture mole average
1, complete data on distributions of M, Tb, S, and n molecular weight is 562. The estimated average mo-
versus mole or weight fractions were available. Mixture lecular weight from eq 5 is 563, while from the gamma
Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997 4301

Figure 1. Comparison of various distribution models for molec- Figure 3. Comparison of various methods for prediction of specific
ular weight of a heavy petroleum mixture. Data are taken from gravity distribution of sample no. 19.
Rodgers et al. (1987).
property, a complete distribution can be determined for
that property.
Now, we present the methods that can generate
property distributions by predicting parameters P0 and
A in eq 1 for different properties with use of only three
mixture bulk properties which are readily available or
measurable. These properties are molecular weight,
specific gravity, and refractive index (at 20 °C) for a C7+
fraction such as those given in Table 1. Two methods
are presented here.

Prediction of Distributions Using Bulk

Properties: M, S, n (Method A)
The beauty of the distribution model given by eq 1
(or eq 4) is that only two parameters (P0 and A) must
be determined for each property in order to predict the
whole distribution, since values of B are known for each
property. For example, if M0 is known, then from
definition of P* we have
Figure 2. Comparison of various models for prediction of boiling
point distribution of sample no. 16. Mav - M0
av ) (13)
and exponential distribution models they are 559 and M0
569, respectively. Similar evaluations for boiling point
and specific gravity distributions for an oil sample (no. where Mav is the mixture bulk molecular weight of the
16) and a condensate sample (no. 19) in Table 1 are C7+ fraction which is assumed to be known from
presented in Figures 2 and 3. For sample no. 19 the measurement. Similarly, S* av and I*
av can be estimated
estimated average specific gravity from eq 5 is 0.7612 from knowledge of S7+ and n7+ (or I7+). Once M*av,
versus the experimental value of 0.7597, giving a S*av, and I*
av are known, AM, AI, and AS can be easily
deviation of 0.2%. The exponential model gives an calculated through eqs 7, 9, and 11 (or 12). However,
average specific gravity of 0.754 with a deviation of as mentioned before, eq 7 was obtained based on a
1.7%. These evaluations show that for very heavy cumulative mole fraction (xm), while eqs 11 and 12 were
fractions eq 1 predicts better distributions than the derived based on a cumulative weight fraction (xw). Once
gamma distribution model especially toward the heavy distributions for M and S are known, distributions of
end. The flexibility of the proposed model is due to the boiling point (Tb) can be determined using equations
role of parameter B in the exponential part of eq 4. This given by Riazi and Daubert (1987):
is the main difference between the proposed model and
the gamma distribution model. The exponential model
Tb ) 3.76587 exp(3.7741 × 10-3M + 2.98404S -
is not appropriate for heavy fractions as shown in Figure
1. Figures 2 and 3 show that the exponential function 4.25288 × 10-3MS)M0.40167S-1.58262 (14)
is not suitable for presentation of boiling point and
specific gravity distributions. While it seems that eq 1 where Tb is in degrees Kelvin. This equation, which was
is equivalent to the gamma distribution model for derived based on data for 140 pure hydrocarbons,
presentation of molecular weight and boiling point predicts boiling points with an average absolute devia-
distributions, eq 1 actually is a two-parameter function tion of 1%. This equation can also be applied to narrow
because parameter B can be fixed for each property. boiling range fractions, but it cannot be applied to crude
Therefore, when parameters A and P0 are known for a oils or C7+ fractions.
4302 Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997

Since parameters M0, S0, and I0 are not known, at 9. Using S distribution, convert weight fractions (xw)
first we should guess some initial values for these into volume fractions (xv).
parameters. For example, values of 72 for M0 and 0.59 10. Using M and S versus cumulative weight frac-
for S0 can be used as the initial guesses as these are tions, calculate Tb for each cut from eq 14.
the lowest values for samples in Table 1. Fortunately, 11. Using M and S, calculate I from eq 16 (or 17).
analysis of data on values of M0, S0, and I0 for samples 12. From data on I versus cumulative volume fraction
given in Table 1 indicates that there is a good relation (xv), calculate parameters I0 and AI from eqs 1 and 3.
among these three parameters as follows: Then calculate Iav from eqs 9 and 10.
13. Calculate 1 ) |(Iav,calc. - I7+)/I7+|.
I0 ) 0.7454 exp(-0.01151M0 - 2.37842S0 + 14. If 1 e 0.005, continue from step 15. Otherwise,
0.01225M0S0)M00.2949S01.53147 (15) go back to step 6 with Snew ) Sold + 0.005 and repeat
steps 7-13.
The above equation can reproduce values of I0 with an 15. Calculate I0 from eq 15.
AAD of 0.3%. However, based on properties of more 16. Calculate 2 ) |(I0,calc. - I0,step12)/I0,calc.|.
than 500 pure compounds and narrow boiling range 17. Go back to step 2 with a new guess for M0 (higher
petroleum fractions, a more general relation for the than the initial guess). Repeat steps 2-17 until 2
prediction of parameter I was obtained in the following becomes minimum and less than 0.005.
form: 18. Using data for Tb versus cumulative weight
fraction, determine parameters T0 and AT from eqs 1
for M e 300 and 3.
19. Print M0, AM, S0, AS, T0, AT, I0, and AI.
I ) 0.12399 exp(3.4622 × 10-4M + 0.90389S - 20. Generate distributions for M, S, Tb, and I using
6.0955 × 10-4MS)M0.02264S0.22423 (16) eq 1 with parameters in step 19.

for M > 300 Prediction of Distributions Using M, S, and TBP

(Method B)
I ) 0.01102 exp(-8.61126 × 10 M + 3.228607S +

9.07171 × 10-4MS)M0.02426S-2.25051 (17) In some cases TBP (true boiling point) distribution
of a sample is known through the distillation curve. For
The above equations predict refractive indices with AAD such samples only two bulk properties such as M and
of 0.05%. Similar relations are available to estimate S or M and n are needed. For these cases, with an
the refractive index from boiling point and specific initial guess on S0, an initial distribution for S can be
gravity as given by Riazi and Daubert (1987) or the API determined. Then using Tb and S for each subfraction,
Technical Data Book (1989): M can be estimated through the relation given by the
API Technical Data Book (1989):
for M e 300
I ) 2.3435 × 10-2 exp(7.029 × 10-4Tb + 2.468S - M ) 42.9654 exp(2.097 × 10-4Tb - 7.787S +

10.267 × 10-4TbS)T0.0572 S-0.72 (18) 2.0848 × 10-3TbS)T1.26007

b S4.98308 (20)

for M > 300 Using eq 18, parameter I and as a result its distribution
can be determined. For this method the calculation
I ) 1.8422 × 10 -2
exp(11.6352 × 10 Tb + 5.144S -
procedure can be summarized as follows:
1. Read values of M and S and the TBP distribution
5.92 × 10-4TbS)T-0.4077 S-3.333 (19) (e.g., simulated distillation curve) for a given crude oil
In these relations Tb is in degrees Kelvin. The refractive 2. Choose 20 arbitrary cuts for the mixture with
index, n, can be estimated through eq 2 with parameter equal weight (or volume) fractions of 0.05. Then find
I estimated from the above relations. values of boiling points at these points from the TBP
Steps to predict distributions for M, Tb, S, and I (or curve. If simulated distillation is available, weight
n) from three bulk properties of C7+ can be summarized fraction should be used for each cut.
as follows: 3. Find parameters T0 and AT using eq 3 and data
1. Read values of M, S, and I for a given C7+ sample. given in step 2.
If I is not available, eq 16 or 17 may be used to estimate 4. Guess an initial value for S0 (assume S0 ) 0.59).
this property. 5. Calculate AS from eq 11 (or 12) using Newton’s
2. Guess an initial value for M0 (assume M0 ) 72) method.
and calculate M* av from eq 13.
6. Find the distribution of S in terms of the cumula-
3. Calculate AM from eq 7. tive weight fraction using eqs 1 and 2. The specific
4. Choose 20 arbitrary cuts for the mixture with gravity can be used to convert volume fractions into
equal mole fractions of 0.05. Then calculate M for each weight fractions if TBP is originally available in terms
cut from eq 1. of the cumulative volume fraction.
5. Convert mole fractions (xm) to weight fractions (xw) 7. Using values of Tb and S for each cut, calculate M
using molecular weights obtained in step 4. from eq 20.
6. Guess an initial value for S0 (assume S0 ) 0.59). 8. Use values of M calculated in step 7 to convert
7. Calculate AS from eq 11 (or 12) using Newton’s weight fractions into mole fractions (xm).
method. 9. Using data calculated in step 8, find the molar
8. Find the distribution of S in terms of cumulative distribution of molecular weight. In this step param-
weight fraction from eq 1 and calculate S for each cut. eters M0 and AM are calculated from eq 3.
Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997 4303
10. Calculate values of I from T and S for each cut Table 2. Gaussian Quadrature Points and Weights for 3
using eq 18 (or 19). and 5 Points (from Abramowitz and Stegun, 1972; p 923)
11. Find values of I0 and AI by eq 3 using data i point y weight wi
generated in step 10. N)3
12. Using M0 obtained in step 9 and S0 assumed in 1 0.415 774 556 783 7.110 930 099 29 × 10-1
step 4, calculate a value for I0 from eq 15. 2 2.294 280 360 279 2.785 177 335 69 × 10-1
13. Assume I1 ) I0 calculated from step 11 and I2 ) 3 6.289 945 082 937 1.038 925 650 16 × 10-2
I0 calculated from step 12. Then calculate I ) |(I2 - N)5
I1)/I1|. 1 0.263 560 319 718 5.217 556 105 83 × 10-1
2 1.413 403 059 107 3.986 668 110 83 × 10-1
14. If I g 0.005, go back to step 4 using S0,new ) S0,old 3 3.596 425 771 041 7.594 244 968 17 × 10-2
+ 0.005. If I < 0.005, go to step 15. 4 7.085 810 005 859 3.611 758 679 92 × 10-3
15. Print T0, AT, M0, AM, S0, AS, I0, and AI. 5 12.640 800 844 276 2.336 997 238 58 × 10-5
16. Generate complete distributions for M, S, Tb, and
I using eq 1 with parameters determined from step 15. is by specifying the range of a property such as molec-
ular weight or boiling point or the range of carbon
Splitting and Lumping Schemes number for each pseudocomponent. For example, a C7+
fraction can be divided into 5 pseudocomponents with
Usually detailed analytical data for reservoir fluids carbon number ranges C7-C10, C11-C15, C16-C25, C26-
are available from C1 to C6 and for compounds heavier C35, and C36+. The lower and upper values of molecular
than C7; they are grouped into a single C7+ fraction with weight for each group can be determined from the
known bulk properties. For such cases method A values given by Riazi and Al-Sahhaf, (1996). The
outlined above can be used to obtain distribution of molecular weight ranges are M0-136, 136-207, 207-
various properties. Once the molar distribution is 345, 345-485, and 485-∞. These ranges are arbitrary
known, then C7+ can be splitted into a number of and may vary according to the number of pseudocom-
pseudocomponents with known mole fractions, molec- ponents chosen. Having the lower and upper limits for
ular weights, and specific gravities. In many cases each group, one can determine the mole fraction and
detailed analytical data on a reservoir fluid are available molecular weight of the group using the probability
to higher carbon numbers and heavy compounds are density function:
grouped into a C20+ or even C30+. For these cases if
molecular weight and specific gravity are available
versus molar composition of each SCN, then eq 1 can zi ) ∫P*P* F(P*) dP*

be conveniently applied to determine molecular weight
and specific gravity distributions. But if M and S data
are not available for each SCN, one may use properties
of SCN groups recommended by Riazi and Al-Sahhaf
P*i,av ) ( )∫
P*F(P*) dP* (24)

(1996) or Katz and Firoozabadi (1978) to obtain complete where i varies from 1 to the total number of pseudocom-
distributions. However, for the sake of the computa- ponents. Pi,av is the average value for property P for
tional efficiency, it is necessary to lump various SCN the pseudocomponent i. Substituting F(P*) from eq 4
groups into a few (3, 5, or 7) pseudocomponents. In into the above integrals, we get the following relations
order to do so, data available on M and S distributions for the property Pi,av and mole fraction zi for the
for SCN groups heavier than C6 should be used to obtain pseudocomponent i.
distribution functions through eq 1 and then pseudocom-
ponents can be determined from distribution functions.
Two methods are recommended in this paper to obtain
these pseudocomponents. The first method is applica-
zi ) exp - P*i-1B - exp - P*i B
A A ) ( ) (25)

( )( ) [ ( ) ( )]
tion of the Gaussian quadrature method as discussed 1 A 1/B 1 1
by Stroud and Secrest (1966). The Gaussian quadrature P*i,av ) Γ 1+ ,qi-1 - Γ 1+ ,qi (26)
zi B B B
method is used to provide a discrete representation of
continuous functions using different numbers of quadra- B
ture points. The number of pseudocomponents is the qi ) P*i B (27)
same as the number of quadrature points. This method
gives the following relations for calculation of properties Pi,av ) P0(1 + P*
i,av) (28)
and mole fractions of the pseudocomponents:
where Γ(1+1/B,qi) is the incomplete gamma function
A 1/B 1/B
[ ()
Pi ) P0 1 +
yi ] (21) which can be evaluated as discussed in various math-
ematical handbooks (e.g., Press et al., 1986; pp 160-
zi ) wi (22) 161) or through computer software such as MATHE-
MATICA. If B ) 1, then eq 26 reduces to
where yi are quadrature points and wi are weighting P*
i,av )
factors. Pi and zi are the values of property P and mole
fractions for the pseudocomponents. Distrbution pa-
rameters P0, A, and B are known from methods dis-
cussed in the previous section. Sets of yi and wi are
( )[(
A ) (
exp -
- 1+
exp -
) ( ) ( )]
tabulated for various numbers of quadrature points (29)
(e.g., see Abramowitz and Stegun, 1972, p 923). Table
2 lists values of yi and wi for 3 and 5 quadrature points. The two methods for splitting and lumping schemes
The second method to obtain the pseudocomponents described above may be summarized as follows:
4304 Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997

Table 3. Evaluation of Proposed Methods for Prediction

of Property Distributions of C7+ Samples in Table 1
(Number of Samples ) 48; Number of Subfractions ) 650)
average absolute deviation (AAD)
M T, K S n
method A 8.1 8 0.005 0.01
method B 4.3 4 0.005 0.01

Method I: Gaussian Quadrature Approach

1. Read composition of SCN groups and properties
of plus fractions (e.g., M20+ and S20+). Normalize the
mole fractions.
2. If M and S for each SCN group are not available,
obtain these properties from Riazi and Al-Sahhaf (1996).
3. Determine distribution parameters for molecular
weight (M0, AM, and BM) in terms of cumulative mole
fraction and for specific gravity (S0, AS, and BS) in terms
of cumulative weight fraction.
4. Choose the number of pseudocomponents or lumped Figure 4. Predicted distributions for molecular weight of sample
groups (e.g., 5) and determine their mole fractions (zi) no. 39.
and molecular weights (Mi) from eqs 21 and 22 using
the quadrature points given in Table 2.
5. Using molecular weights and mole fractions from
step 4, determine discrete weight fractions (zwi) for each
6. Convert discrete weight fractions into cumulative
weight fractions and then use eq 1 with parameters S0
and AS from step 3 to obtain the specific gravity for each
pseudocomponent. For example, S1 (specific gravity of
the first pseudocomponent) can be determined directly
from eq 1 at x1 ) zw1/2.
7. Obtain Mav and Sav for the mixture from Mav )
∑ziMi and 1/Sav ) ∑zwi/Si.

Method II: Carbon Number Range Approach

1. Same as method I.
2. Same as method I.
3. Same as method I.
4. Choose the number of pseudocomponents (e.g., 5)
and the carbon number ranges: C7-C10, C11-C15, C16- Figure 5. Predicted distributions for boiling point of sample no.
C25, C26-C35, C36+. 39.
5. Obtain molecular weights for the carbon number
ranges chosen in step 4 from Riazi and Al-Sahhaf (1996). nearly the same distributions with average deviations
The first molecular weight is M0. The molecular weight of about 0.6%. However, in using method B, since the
ranges are: M0-136, 136-207, 207-345, 345-485, and TBP distribution was used, more accurate predictions
485-∞. for molecular weight and especially boiling point were
6. Obtain mole fractions of pseudocomponents (zi) obtained. Figures 4-7 show graphical evaluations with
from eq 25 and molecular weights (Mi) from eqs 26-28 actual data for sample no. 39 whose properties are given
or eq 29 if BM ) 1. in Table 1. The distribution of molecular weight versus
7. Same as step 5 in method I. cumulative mole fraction shown in Figure 4 is also
8. Same as step 6 in method I. presented in the form of a probability density function
9. Same as step 7 in method I. determined by eq 4 as shown in Figure 8. Results
In this method if the calculated mole fraction for a presented in Table 3 and Figures 4-8 indicate that both
pseudocomponent (in step 6) is too high or too low, we methods proposed here are capable of generating com-
may reduce or increase the carbon number range chosen plete distributions for the four basic characterization
for that pseudocomponent in step 4. An example for parameters, namely, M, Tb, S, and I. Once these basic
use of these two methods is presented in the next parameters are determined, then correlations of Riazi
section. and Daubert (1980, 1987) can be used to estimate
critical parameters needed in equations of state calcula-
Results and Discussion tions.
To show an example for the lumping procedures as
For samples listed in Table 1 parameters M0, AM, T0, described in the previous section, we used analytical
AT, S0, AS, I0, and AI were predicted by both methods A data available for sample no. 42 in Table 1. Molecular
and B. Errors produced for distribution of various weight and specific gravities are available from C7 to
properties for all 48 samples are given in Table 3. As C20+ for 14 groups. Methods I and II described for the
is shown by the results in this table, for specific gravity lumping and splitting procedures were used to obtain
and refractive index both methods A and B produced 5 pseudocomponents. Mole fractions, molecular weights,
Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997 4305

Figure 6. Predicted distributions for specific gravity of sample Figure 8. Predicted probability density function for molecular
no. 39. weight of sample no. 39.

probability density functions in terms of cumulative

mole fraction for the boiling points of the feed, liquid,
and vapor streams, respectively, then from material
balance we have

FF(T) ) (1 - φ)FL(T) + φFV(T) (30)

in which φ is the mole ratio of vapor product to the feed.
If the system is assumed to be ideal, then

[FV(T) dT]p ) [FL(T) dT]p*(T,TS) (31)

where p* is the vapor pressure function for the fraction

whose boilng point is T and its mole fraction in the vapor
phase is FV(T) dT. TS is the temperature at which
distillation occurs. The above relation can be written

FV(T) p ) FL(T) p*(T,TS) (32)

Figure 7. Predicted distributions for refractive index parameter
of sample no. 39. For all three probability density functions, FF, FV, and
FL we have
and specific gravities of these pseudocomponents are
given in Table 4. Mixture average molecular weight and
specific gravity calculated from the pseudocomponents
∫T∞FFT dT ) ∫T∞FVT dT ) ∫T∞FLT dT ) 1
0 0 0
are very close to the experimental values with devia-
tions of about 0.1%. From eqs 30, 32, and 33 one can derive the following
For further evaluation of proposed methods, experi- relation for calculation of parameter φ:
mental data on boiling point distribution for a Russian
p - p*(T,T )
crude oil as given by Ratzsch et al. (1988) were also
used. In this case TBP distributions for feed, vapor, and
∫T∞(1 - φ)p + φp*(T,T
FF dT ) 0
) T
liquid streams during flash distillation of the crude are
available. Molecular weight, specific gravity, and re- Combination of Trouton’s rule for the heat of vaporiza-
fractive index of the mixture are 220, 0.8334, and tion and the Clasius-Clapeyron equation leads to the
1.4626, respectively. If FF(T), FL(T), and FV(T) are the following relation for the vapor pressure:
Table 4. Lumping of 14 Fractions from C7 to C20+ for Sample No. 42 into Five Pseudocomponents by Two Methodsa
method I: Gaussian quadrature approach method II: carbon number range approach
pseudocomponent i mol. frac. zi wt. frac. zwi Mi Si mol. frac. zi wt. frac. zwi Mi Si
1 0.5218 0.3493 102.1 0.7436 0.532 0.372 106.7 0.7457
2 0.3987 0.4726 180.8 0.8023 0.302 0.328 165.5 0.7957
3 0.0759 0.1645 330.4 0.8591 0.144 0.240 254.4 0.8389
4 0.0036 0.0134 569.5 0.9174 0.019 0.049 392.7 0.8847
5 2.3 × 10-5 1.42 × 10-4 950.1 0.9809 0.003 0.011 553.5 0.9214
mixture 152.5 0.7905 152.5 0.7908
a Experimental C
7+ properties: M7+ ) 151.6; S7+ ) 0.7917. Distribution parameters: M0 ) 84; AM ) 0.7157; BM ) 1. For specific
gravity in terms of the cumulative weight fraction: S0 ) 0.655, AS ) 0.038 75, BS ) 3.
4306 Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997

TBP distributions were in good agreement with the

experimental data. Finally splitting and lumping
schemes of C7+ samples for the proposed distribution
model are presented through application of the Gauss-
ian quadrature technique or specification of the molec-
ular weights of pseudocomponents. Estimated proper-
ties are in excellent agreement with the experimental

A, B ) constants in eq 1
C1, C2 ) regression coefficients in eq 3
d ) density, g/cm3
F ) probability density function given by eq 4
I ) refractive index parameter defined in eq 2
M ) molecular weight
n ) refractive index at 20 °C and 1 atm
P ) property such as M, Tb, S, I, and d
Figure 9. Predicted density functions of feed, liquid, and vapor p ) pressure
at 300 °C for flash vaporization of a Russian crude oil. Actual data p* ) vapor pressure
are taken from Ratzsch et al. (1988). q ) parameter defined in eq 27
S ) specific gravity at 15.5 °C (60 °F)
p*(T,TS) ) pa exp[10.58(1 - T/TS)] (35) T ) boiling point in distribution functions, K
Tb ) normal boiling point, K
where pa is the atmospheric pressure. In this relation TS ) temperature at which distillation occurs, K
T is the boiling point of each cut in the distribution X ) regression parameter defined in eq 3
model. By combining eqs 30 and 32, we can get x ) cumulative volume, weight, or mole fraction
p Y ) regression parameter defined in eq 3
FL(T) ) FF(T) (36) yi ) Gaussian quadrature points
(1 - φ)p + φp*(T,TS) zi ) mole fraction of pseudocomponent i
zwi ) weight fraction of pseudocomponent i
FV(T) ) FF(T) (37) wi ) Gaussian quadrature weighting factors
(1 - φ)p + φp*(T,TS)
Greek Symbols
The proposed method has been applied to generate FF-  ) error parameter
(T) and subsequently FL(T) and FV(T) from eqs 36 and φ ) vapor to feed mole ratio in single-stage flash distillation
37, respectively. Parameters T0, AT, and BT for the feed γ ) activity coefficeint of species in the mixture
which were obtained from method A are 241.7 K, 1.96,
and 3, respectively. Results for the probability density Superscripts
functions are compared with the actual data (Ratzsch * ) dimensionless parameter defined in eq 1 for property
et al., 1988) in Figure 9. Results for flash distillation P as P* ) (P - P0)/P0
presented in Figure 9 are at 300 °C and 1 atm. Under F ) feed in a distillation unit
these conditions parameter φ was calculated as 0.79 in L ) liquid product in a distillation unit
which the actual value was 0.83. Part of errors for V ) vapor product in a distillation unit
predicted distributions of FL(T) and FV(T) is due to
assumption of an ideal solution for VLE calculations as Subscripts
well as an approximate relation for the estimation of 0 ) initial value for any property at x ) 0
vapor pressures. For more accurate calculations, the a ) atmospheric pressure
activity coefficient parameter can be included in the av ) average bulk property for the mixture
above relations by replacing p*(T,TS) with γ(T) p*(T,TS) I ) refractive index parameter
where γ(T) is the activity coefficient for species whose i ) properties for a SCN group i
boiling point is T, and it may be estimated through i,av ) average property for pseudocomponent i
methods such as the UNIFAC model. M ) molecular weight parameter
T ) boiling point parameter
Conclusions S ) specific gravity parameter
In this work, based on a two-parameter distribution
model, a method is presented to predict complete Literature Cited
distributions for basic characterization parameters of Abramowitz, M.; Stegun, I. A. Handbook of Mathematical Func-
a hydrocarbon-plus fraction. According to the proposed tions; Dover Publications, Inc.: New York, 1972.
method, when three bulk properties (M, S, n) are Ahmed, T. Hydrocarbon Phase Behavior; Gulf Publishing Co.:
available for a C7+ fraction, distribution functions for Houston, 1989.
M, Tb, S, and I can be predicted with good accuracy. API Technical Data BooksPetroleum Refining, 5th ed.; Daubert,
However, distributions can be predicted more accu- T. E., Danner, R. P., Eds.; American Petroleum Institute
(API): Washington, DC, 1989; Chapter 2, pp 15-30.
rately, if in addition to TBP data M and S are available. Austad, T.; Hvidsten, J.; Norvik, H.; Whitson, C. H. Practical
The method has been used to predict boiling point Aspects of Characterization Petroleum Fluids. Presented at the
distributions of feed, vapor, and liquid products in flash Conference on North Sea Condensate Reservoirs and Their
distillation of a Russian crude oil sample, and predicted Development, London, May 24-25, 1983.
Ind. Eng. Chem. Res., Vol. 36, No. 10, 1997 4307
Berge, O. Damp/Vaeske-Likevekter i Raoljer: Karakterisering av Ratzch, M. T.; Kehlen, H.; Schumann, J. Flash Calculations for a
Hydrokarbonfraksjon. M.Sc. Thesis, Department of Chemical Crude Oil by Continuous Thermodynamics. Chem. Eng. Com-
Engineering, Norwegian Institute of Technology, Trondheim, mun. 1988, 71, 113-125.
Norway, 1981. Riazi, M. R. Distribution Model for Properties of Hydrocarbon-
Haaland, S. Characterization of North Sea Crude Oils and Plus Fractions. Ind. Eng. Chem. Res. 1989, 28, 1831-1735.
Petroleum Fractions. M.Sc. Thesis, Department of Petroleum Riazi, M. R.; Daubert, T. E. Simplify property predictions. Hydro-
Engineering, Norwegian Institute of Technology (NTH), Trond- carbon Process. 1980, 3, 115-116.
heim, Norway, 1981. Riazi, M. R.; Daubert, T. E. Characterization Parameters for
Hariu, O. H.; Sage, R. C. Crude Split Figured by Computer. Petroleum Fractions. Ind. Eng. Chem. Res. 1987, 26, 755-759.
Hydrocarbon Process. 1969, 143-148. Riazi, M. R.; Al-Sahhaf, T. A. Physical Properties of Heavy
Jacoby, R. H.; Berry, V. J. A Method for Predicting Pressure Petroleum Fractions and Crude Oils. Fluid Phase Equilib. 1996,
Maintenance Performance for Reservoirs Producing Volatile 117, 217-224.
Crude Oil. Trans. AIME 1958, 213, 59-64. Rodgers, P. A.; Creagh, A. L.; Prange, M. M.; Prausnitz, J. M.
Katz, D. L.; Firoozabadi, A. Predicting Phase Behavior of Con- Molecular Weight Distributions for Heavy Fossil Fuels from Gel-
densate/Crude-Oil Systems Using Methane Interaction Coef- Permeation Chromatography and Characterization Data. Ind.
ficients. JPT, Trans. AIME 1978, 1649-1655. Eng. Chem. Res. 1987, 26, 2312-2318.
Kawanada, S.; Park, S. J.; Mansoori, G. A. Organic Deposition Soreide, I. Improved Phase Behavior Predictions of Petroleum
from Reservoir Fluids: A Thermodynamic Predictive Technique. Reservoir Fluids from A Cubic Equation of State. Doktor Ing.
SPE Reservoir Eng. 1991, May, 185-192. Thesis, Department of Petroleum Engineering, Norwegian
Lee, S. T.; Jacoby, R. H.; Chen, W. H.; Culham, W. E. Experimental Institute of Technology (NTH), Trondheim, Norway, 1989.
and Theoretical Studies on the Fluid Properties Required for Stroud, A. H.; Secrest, D. Gaussian Quadrature Formulas; Pren-
Simulation of Thermal Processes. The SPE 54th Annual Fall tice-Hall: Englewood Cliffs, NJ, 1966.
Technical Conference and Exhibition, Las Vegas, Sept 23-26, Whitson, C. H. Characterizing Hydrocarbon Plus Fractions. Soc.
1979; Paper SPE 8389. Pet. Eng. J. 1983, Aug, 683-694.
Pedersen, K. S.; Thomassen, P.; Fredenslund, A. Thermodynamics Whitson, C. H. Effect of C7+ Characterization on Equation of State
of Petroleum Mixtures Containing Heavy Hydrocarbons. 1. Predictions. Soc. Pet. Eng. J. 1984, May, 685-696.
Phase Envelope Calculations by Use of the Soave-Redlich- Whitson, C. H.; Anderson, T. F.; Soreide, I. C7+ Characterization
Kwong Equation of State. Ind. Eng. Chem. Process Des. Dev. of Related Equilibrium Fluids Using the Gamma Distribution.
1984a, 23, 163-170. AIChE Spring Meeting, New Orleans, March 6-10, 1988.
Pedersen, K. S.; Thomassen, P.; Fredenslund, A. Thermodynamics
of Petroleum Mixtures Containing Heavy Hydrocarbons. 2. Received for review April 3, 1997
Flash and PVT Calculations with the the SRK Equation of
Revised manuscript received June 26, 1997
State. Ind. Eng. Chem. Process Des. Dev. 1984b, 23, 566-573.
Pedersen, K. S.; Thomassen, P.; Fredenslund, A. Thermodynamics
Accepted July 1, 1997X
of Petroleum Mixtures Containing Heavy Hydrocarbons. 3. IE970260Y
Efficient Flash Calculation Procedures Using the SRK Equation
of State. Ind. Eng. Chem. Process Des. Dev. 1985, 24, 948-954.
Press, W. H.; Flannery, B. P.; Teukolsky, S. A.; Vetterling, W. T.
X Abstract published in Advance ACS Abstracts, September
Numerical Recipes, The Art of Scientific Computing; Cambridge
University Press: Cambridge, London, 1986. 1, 1997.